gloomy

An n-gram database written in Go, optimized for write once read many use.

• Building an index
• Searching
• Config reference
• Advanced source data filtering
• Additional functions

Building an index

Gloomy supports the following text formats (specified via sourceType conf. value):

• vertical format.
• plain text files

gloomy -ngram-size 3 create-index ./config.json

where config.json looks like this:

{
    "inputFilePath": "/path/to/a/vertical/file",
    "sourceType": "vertical",
    "filterArgs": [],
    "ngramIgnoreStructs": [],
    "ngramStopStrings": [".", ":"],
    "ngramIgnoreStrings": ["\"", ","],
    "tmpDir": "/tmp/gloomy",
    "procChunkSize": 1000000,
    "outDirectory": "/path/to/an/output/directory",
    "args": {
      "doc.file": "col8",
      "doc.n": "col8",
      "head.type": "col8"
  }
}

Searching

In the searching mode, a gloomy.conf file (by default in the working directory) is expected:

{
    "dataPath": "/path/to/indices/data",
    "serverPort": 8090,
    "serverAddress": "127.0.0.1"
}

command line mode

gloomy search corpname phrase

HTTP server mode

Start a server:

gloomy search-service

Test a client:

curl -XGET http://localhost:8090/search?corpus=susanne&q=from

Query syntax

The current version supports only a search by the first token.

Exact search:

gloomy search susanne absolute

... searches for all the n-grams with the first token equal to absolute.

Search by a prefix:

gloomy search susanne abs*

... searches for all the n-grams where the first token starts with abs*

Search by a regular expression:

gloomy search -qtype regexp susanne "dogs?"

Please note that Gloomy's support of regular expressions is limited:

1. . (dot), [abc], a?, a.*, a+, (foo)
2. no character groups (e.g. \w, \s)
3. alternation (the | operator) behaves differently - it has the highest priority:
   • foo|bar translates into either fooar or fobar
   • use (foo)|(bar) to get either foo or bar

Metadata retrieval

Command line:

gloomy search --attrs doc.file,doc.n susanne absolute

In HTTP server mode use multi-value attribute:

http://localhost:8090/search?corpus=susanne&q=from&attrs=doc.file&attrs=doc.n

http://localhost:8090/search?corpus=susanne&qtype=regexp&q=dogs%3F&attrs=doc.file&attrs=doc.n

Config reference

inputFilePath - path to a source file in a plain text or zipped plain text format

sourceType - plain/vertical

filterArgs - a CNF encoded set of rules applied to structural attributes a a filter

ngramIgnoreStructs - a list of structs to ignore

ngramStopStrings - a list of strings to end an n-gram (typically: ".", "!" etc.)

ngramIgnoreStrings - a list of strings to be completely ignored

tmpDir - a directory where Gloomy may store temporary data when dealing with large data; the directory may not exist - Gloomy will create it if needed

procChunkSize - number of ngrams per temporary chunk file when dealing with large data

outDirectory - output directory

args - structural attributes to be imported

Advanced source data filtering

To filter specific ngrams out Gloomy offers a way how to call a custom external function testing current n-gram (and its possible PoS tag companion).

package main

import (
	"regexp"
	"github.com/tomachalek/gloomy/index/builder/filter"
)

var (
    tagPattern = regexp.MustCompile("^.{14}8.")
)

func filterF1(words []string, tags []string) bool {
	return !tagPattern.MatchString(tags[i])
}

var FilterF1 = filter.CustomFilter(filterF1)

Compile the function(s) with

go build -buildmode=plugin

Then upgrade your config json:

{
    "inputFilePath": "/path/to/a/vertical/file",
    "...": "...",
    "ngramFilter": {
      "lib": "/path/to/your/filter.so",
      "fn": "FilterF1"
    },
}

Additional functions

Extracting sorted unique n-grams with frequencies

It is possible to just extract n-grams to a raw text file instead of building an index:

gloomy -ngram-size 3 extract-ngrams ./config.json